摘要
低光照图像噪声大、颜色失真、能见度低,这些问题对低光照环境下计算机视觉任务带来挑战。提出了一种基于双支路特征的生成对抗网络(DBF-GAN)进行低光照图像增强。该模型采用非配对数据集训练,模型中生成器设计成两个分支特征网络,即全局特征信息网络和全局特征像素网络。两个网络获得的特征图在经过STA(Structure-Texture Aware Module)模块进行增强,同时设计一个多尺度判别器引导生成器增强图像。经过与多种算法对比实验表明,该算法增强结果的结构相似度(PSNR)和峰值信噪比(SSIM),分别是21.181和0.857,均优于所对比的算法。
In this paper,a dual branch feature based on generation adversarial network(DBF-GAN)is proposed for low-light image enhancement.The network is trained with unpaired datasets.In the network,the generator is designed into two branch networks:global feature information network and global feature pixel network.The feature maps obtained by the two branch networks are enhanced by structure texture aware(STA)module.Meanwhile,a multiscale discriminator is designed to guide the generator to enhance images.Compared with other algorithms,the experimental results indicate that the structure similarity(SSIM)and peak signal-to-noise ratio(PSNR)of our algorithm are 21.181 and 0.857 respectively.
出处
《工业控制计算机》
2022年第12期40-42,共3页
Industrial Control Computer
关键词
生成对抗网络
双支路特征
低光照图像
图像增强
generative adversarial networks
dual branch feature
low light images
image enhancement